Using cheap model Jev as a code rubric reviewer to fix agent slop code, 100x cheaper than CodeRabbit
Nedomas · reddit · 2026-09-18
The author's AI software factory was producing increasingly sloppy code, so they tried using the newly released cheap model Jev to steer agents out of slop loops.
What worked:
- Sending the codebase to Jev for rating works with LLMs in general, but is expensive with big models;
- Jev was bad at first, but the author built better rubrics — such as detecting specific bad-code patterns — and it finally clicked;
- The author claims results are now on par with CodeRabbit-level review at roughly 100x lower cost.
The post ends by asking the community for tricks and patterns on using Jev for coding tasks.
More from coding & agent
- 'Jev' agent plays Pac-Man with Astra strategizing and millisecond execution — daniel_mac8 · 2026-09-18
- Awesome Code-as-X: curated papers unifying agent research via programs as representations — ccloy · 2026-09-18
- 8 AI personal agents, zero successful SMS — 10DLC is the blocker — iamstanty · 2026-09-18
- Zalando details its Kubernetes-based internal agent platform with kagent and kro — bibryam · 2026-09-18
- Modern coding: fantasize a crazy concept on Twitter, then hand the link to a coding agent — airesearch12 · 2026-09-18
- Dev swapped to Opus 5 after OpenAI credits ran out — 'this is not great' — lucasmeijer · 2026-09-18